davidjosipovic/news-trend-analysis
Automated NLP pipeline for news analysis with sentiment detection, topic modeling & summarization. Daily updates via GitHub Actions, 24/7 Streamlit dashboard on Railway.
What's novel
Automated NLP pipeline for news analysis with sentiment detection, topic modeling & summarization. Daily updates via GitHub Actions, 24/7 Streamlit dashboard on Railway.
Code Analysis
10 files read · 3 roundsAn automated news aggregation pipeline that fetches economic articles, performs sentiment analysis (RoBERTa), topic modeling (BERTopic), summarization (DistilBART), and predictive analytics (XGBoost/ElasticNet) with a Streamlit dashboard and FastAPI endpoint.
Strengths
The predictive analytics layer is genuinely well-engineered with proper OOP design, time-based splits, SMOTE for imbalance, dual-model comparison, and thoughtful feature engineering including domain-specific Croatian holidays. The semantic duplicate detection using sentence-transformer embeddings with cosine similarity is a non-trivial addition that goes beyond simple URL deduplication.
Weaknesses
The README claims FinBERT but the code uses cardiffnlp/twitter-roberta-base-sentiment-latest, and the test suite appears minimal (only 4 test files for a 31-source-file project). The pipeline is tightly coupled to file paths and CSV intermediates rather than using a proper data pipeline framework, making it fragile and hard to test in isolation.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
328
Contributors
2
Files
368
Active weeks
20
Repository
Language
Python
Stars
2
Forks
2
License
—